Suscetibilidade a escorregamentos na bacia hidrográfica do médio/alto Rio Taquari-Antas, RS: utilização de técnicas de machine learning

• Autor(a) principal: Sampaio, Francisco Monte Alverne de Sales
• Data de Publicação: 2024
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Manancial - Repositório Digital da UFSM
• dARK ID: ark:/26339/001300000vf5m
• Texto Completo: http://repositorio.ufsm.br/handle/1/31862
• topic: Desastres naturais; Escorregamentos; Machine learning; Natural disasters; Slides; CNPQ::CIENCIAS HUMANAS::GEOGRAFIA

Planar slides are a landslide type that can cause natural disasters with economic impacts and loss of lives. The increase in these events is associated with population growth and unplanned urbanization. In Brazil, slides resulted in 3,758 deaths between 1988 and 2022. Mapping planar slide susceptibility is vital for prevention and mitigation of these disasters, and machine learning techniques, such as the Maximum Entropy Model (MAXENT), have enabled the analysis and manipulation of large volumes of data, producing fast and highly accurate results, becoming a valuable tool to minimize damage in slide-prone areas. This study aimed to map slide susceptibility in the hydrographic basin of the medium/high Taquari/Antas River (SMARTA) using MAXENT. The work was divided into five stages: i) literature review, ii) organization of the cartographic database, iii) identification of scars, iv) identification of conditioning factors, and v) mapping slide susceptibility. To identify planar slide scars between 2000 and 2022, two methods were used: the first involved research in newspapers with defined criteria and systematic data collection, and the second used the visual interpretation of satellite images available in the Google Earth Pro software. Subsequently, the analysis of slide conditioning factors in SMARTA was carried out using the following information plans: i) slope; ii) distance to first-order rivers; iii) distance to highways and secondary roads; iv) distance to structural lineaments, and v) curvature of the hillslopes. To map slide susceptibility, the MAXENT machine learning model was used. Input data consisted of points with slide scars identified visually in Google Earth Pro. The results showed that MAXENT had a global accuracy above 0.94, and frequency ratio indicated a higher occurrence of scars in areas of high and very high susceptibility. Analysis of newspaper and image data revealed 119 scars and one death between 2010 and 2022, and slope was the main conditioning factor for slides in SMARTA. Approximately 1.3% of the SMARTA area was classified as very high susceptibility, mainly in valleys and slopes. The municipality of Caxias do Sul had the largest area classified as very high susceptibility, followed by the municipalities of Bento Gonçalves, Veranópolis, Flores da Cunha, and Campestre da Serra. Finally, the high potential of the MAXENT model for planar slide susceptibility mapping is emphasized.